I buildSYSTEMS THATTHINK.
I design and ship systems that think — live AI support, queue-based infrastructure, and multi-tenant platforms built to scale without breaking.
Your systems, built to think. Live AI, durable infrastructure, and platforms that don't buckle under scale.
Projects that moved the needle.
Live projects, shipped.
Notes on building.
Building a live chat platform with AI handoff
A confidence score under 0.6 pages a human, and the agent inherits the full transcript instead of starting cold.
Deploying a Next.js monorepo on a VPS with PM2
Three Next.js apps on one box, each a named PM2 process, with graceful reloads that drop zero requests.
Intent routing for support chatbots
A small classifier in front of the model picks one of six domains, and reliability jumps because the prompt stops guessing.
Persisting AI conversations: schema and session lifecycle
Messages are immutable rows, a rolling summary caps the context window, and a session closes after thirty idle minutes.
Rate limiting an Express API the practical way
A Redis token bucket keyed by API key, with limit headers on every response and a Retry-After when you say no.
BullMQ in production: retries, backoff, and dead-letter queues
Exponential backoff for transient failures, a poison-message guard, and a failed set you actually watch.
BetterAuth SSO with httpOnly cookies across subdomains
One login at app.fuyad.dev carries to admin.fuyad.dev through a cookie scoped to the parent domain.
Fixing Next.js build heap OOM (NODE_OPTIONS deep-dive)
A build that died at 1.6 GB on a 2 GB box, traced to the default V8 heap cap and fixed without buying more RAM.
Designing a multi-tenant schema with Prisma and PostgreSQL
A shared schema with a tenant_id on every row, enforced by row-level security so a missing filter cannot leak data.
S3 attachments done right: signed uploads and cleanup
Browsers upload straight to S3 with a presigned URL, and a nightly sweep deletes objects no row points to.
WebSocket live sessions with Socket.io: presence and reconnects
Presence keyed by user with a reference count, plus a Redis adapter so two server instances share one room.
Server Actions vs API routes: when to use which
Server Actions for form mutations that return to the same UI, route handlers for anything a third party calls.
Streaming LLM responses to a React UI
Tokens arrive over a ReadableStream and append to state, so the answer types out instead of landing as a wall.
Caching strategy for a dashboard: Redis and revalidation
Expensive aggregates cached in Redis with a short TTL, and a targeted bust on write so numbers stay current.
From Figma to code: a reliable MCP workflow
Pulling design context through MCP turns a screenshot guessing game into exact tokens, spacing, and variants.
Questions, answered.
Full stack web apps with real backends: internal tools, dashboards, CRMs, multi-tenant platforms, and AI features like live support and intent routing. I lean toward systems that need to hold up in production, not throwaway demos.
Let's build
SOMETHING REAL.
Have a system that needs to think, scale, or just finally get shipped? Let's make it happen.
Based in Dhaka, Bangladesh · Available worldwide